GNU bug report logs - #79990
[PATCH v1] Add binary format specifications '%b' and '%B'

Previous Next

To reply to this bug, email your comments to 79990 AT debbugs.gnu.org.
There is no need to reopen the bug first.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
To: bug-gnu-emacs <at> gnu.org
Subject: [PATCH v1] Add binary format specifications '%b' and '%B'
Date: Thu, 11 Dec 2025 14:42:44 -0500

[Message part 1 (text/plain, inline)]

Tags: patch

Tags: patch

Hey everyone,

For quick things in elisp I often use a small helper function to
format binary digits.  C23 now recognizes ‘%b’ and ‘%B’ for the
‘printf’ functions, and some Lisps have similar directives (e.g. ‘~b’
and ‘~B’ in CL, Guile Scheme), so it would be nice to have this in
‘format’.  Attached is a first draft of a patch achieving this.

[v1-0001-Add-binary-format-specifications-b-and-B.patch (text/patch, attachment)]

[Message part 3 (text/plain, inline)]

Thanks,

-- 
Jacob S. Gordon
jacob.as.gordon <at> gmail.com
Please avoid sending me HTML emails and MS Office documents.
https://useplaintext.email/#etiquette

Message #8 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Eli Zaretskii <eliz <at> gnu.org>
To: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>,
 Paul Eggert <eggert <at> cs.ucla.edu>
Cc: 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Fri, 12 Dec 2025 10:04:04 +0200

> From: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
> Date: Thu, 11 Dec 2025 14:42:44 -0500
> 
> For quick things in elisp I often use a small helper function to
> format binary digits.  C23 now recognizes ‘%b’ and ‘%B’ for the
> ‘printf’ functions, and some Lisps have similar directives (e.g. ‘~b’
> and ‘~B’ in CL, Guile Scheme), so it would be nice to have this in
> ‘format’.  Attached is a first draft of a patch achieving this.

Thanks.

If this relies on C23-compliant libc implementation of printf (as I
think it does), then I don't think we want to have a feature in Emacs
which can be used on a small subset of platforms.  We need instead
either to provide our own implementation of that or maybe use Gnulib
printf.

Adding Paul for discussing the possible implementations and the need.

> +@item %b
> +@item %B
> +@cindex integer to binary
> +Replace the specification with the base-two or binary representation of
> +an integer.  Negative integers are formatted in a platform-dependent
> +way.

Don't we want the negative values to be formatted the same on all
platforms?  And if we leave this platform-dependent, then how will
tests you add work on the platforms where the result is different from
what the tests assume?

>       @samp{%b} uses lower case and @samp{%B} uses upper case in the
> +alternate display form prefixes @samp{0b} and @samp{0B}, respectively.
> +The object can also be a floating-point number that is formatted as an
> +integer, dropping any fraction.

Does this support bignums?

Message #11 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
To: Eli Zaretskii <eliz <at> gnu.org>, Paul Eggert <eggert <at> cs.ucla.edu>
Cc: 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Fri, 12 Dec 2025 11:23:59 -0500

On 2025-12-12 03:04, Eli Zaretskii wrote:
> If this relies on C23-compliant libc implementation of printf (as I
> think it does), then I don't think we want to have a feature in
> Emacs which can be used on a small subset of platforms.

Oh you’re right, the fixnum path does call out to sprintf.  I agree
this should be withheld unless it can be made sufficiently portable.

>> +an integer.  Negative integers are formatted in a platform-dependent
>> +way.
>
> Don't we want the negative values to be formatted the same on all
> platforms?  And if we leave this platform-dependent, then how will
> tests you add work on the platforms where the result is different from
> what the tests assume?

I took this warning from the %o and %x descriptions, which seems to be
associated with the ‘binary-as-unsigned’ option.  IIUC the default
behaviour is portable, so I think these statements could be sharpened
or combined into a paragraph about ‘binary-as-unsigned’.  Is this
correct Paul?

>>       @samp{%b} uses lower case and @samp{%B} uses upper case in the
>> +alternate display form prefixes @samp{0b} and @samp{0B}, respectively.
>> +The object can also be a floating-point number that is formatted as an
>> +integer, dropping any fraction.
> Does this support bignums?

Yes, the bignum path calls bignum_to_c_string -> gmp_lib::mpz_get_str
with base 2.

   (let ((x (1- (expt 2 128)))) (and (bignump x) (format "%#B" x)))

Best,

--
Jacob S. Gordon
jacob.as.gordon <at> gmail.com
Please avoid sending me HTML emails and MS Office documents.
https://useplaintext.email/#etiquette

Message #14 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Eli Zaretskii <eliz <at> gnu.org>, "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
Cc: 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Fri, 12 Dec 2025 10:09:01 -0800

>>  				   + ((p[signedp] == '0'
>>  				       && (p[signedp + 1] == 'x'
>> -					   || p[signedp + 1] == 'X'))
>> +					   || p[signedp + 1] == 'X'
>> +					   || p[signedp + 1] == 'b'
>> +					   || p[signedp + 1] == 'B'))

This looks wrong, as "0b" and "0B" can be produced by hexadecimal 
formats like "%.2X". Better would be to replace the entire expression 
with (!float_conversion && p[signedp] == '0' && p[signedp + 1] == 
conversion), perhaps with a comment explaining that "== conversion".


> We need instead
> either to provide our own implementation of that or maybe use Gnulib
> printf.

Gnulib printf would bring in a lot of stuff: converting floating point 
numbers portably, doing multibyte conversion Gnulib style, etc. Although 
it would work, it's overkill.

Instead, I suggest changing styled_format's 'b' and 'B' processing to 
convert fixnums to bignums and then "goto bignum_arg". On nicer 
platforms where plain printf works (e.g., glibc 2.35+) this step can be 
omitted as an optimization, but that's icing on the cake and could be 
done in a later patch.

Message #17 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, Eli Zaretskii <eliz <at> gnu.org>
Cc: 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Fri, 12 Dec 2025 12:14:42 -0800

On 2025-12-12 08:23, Jacob S. Gordon wrote:
> On 2025-12-12 03:04, Eli Zaretskii wrote:
>>> +an integer.  Negative integers are formatted in a platform-dependent
>>> +way.
>>
>> Don't we want the negative values to be formatted the same on all
>> platforms?  And if we leave this platform-dependent, then how will
>> tests you add work on the platforms where the result is different from
>> what the tests assume?
> 
> I took this warning from the %o and %x descriptions, which seems to be
> associated with the ‘binary-as-unsigned’ option.  IIUC the default
> behaviour is portable, so I think these statements could be sharpened
> or combined into a paragraph about ‘binary-as-unsigned’.  Is this
> correct Paul?

Sounds right: %b, %o and %x are all in the same boat.

Come to think of it, how about if we remove binary-as-unsigned instead? 
In practice that experimental variable seems to cause more trouble than 
it hurts. The only response we got[1] to our request for why it's useful 
was actually more an illustration why it's harmful, as the code using it 
mishandled some negative numbers[2].

[1]: https://debbugs.gnu.org/cgi/bugreport.cgi?bug=32252#167
[2]: https://debbugs.gnu.org/cgi/bugreport.cgi?bug=32252#174

Message #20 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
To: Paul Eggert <eggert <at> cs.ucla.edu>, Eli Zaretskii <eliz <at> gnu.org>
Cc: 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v2] Add binary format specifications '%b' and
 '%B'
Date: Thu, 18 Dec 2025 01:31:23 -0500

[Message part 1 (text/plain, inline)]

On 2025-12-12 13:09, Paul Eggert wrote:
> This looks wrong, as "0b" and "0B" can be produced by hexadecimal
> formats like "%.2X". Better would be to replace the entire
> expression with
> (!float_conversion && p[signedp] == '0' && p[signedp + 1] == conversion),
> perhaps with a comment explaining that "== conversion".

Good catch, thanks.  I’ve used this expression in v2 and added a
comments here, and in the bignum section to express the intent.

> Instead, I suggest changing styled_format's 'b' and 'B' processing
> to convert fixnums to bignums and then "goto bignum_arg".

This makes sense, I’ve made this the default in v2.  AFAICT the
existing bignum APIs assert numbers are in the fixnum range, so I had
to expose arguments to skip them.  Is there a better way?

> On nicer platforms where plain printf works (e.g., glibc 2.35+) this
> step can be omitted as an optimization

I took a stab at this in v2 with an autoconf macro, which checks a
small program for the binary flags, e.g.

  #include <stdio.h>
  int main() {
      return printf("%b", 255) != 8;
  }

This seemed like less of a maintenance burden than tracking down
library compliance.  Is this on the right track?

On 2025-12-12 15:14, Paul Eggert wrote:
> Come to think of it, how about if we remove binary-as-unsigned
> instead? In practice that experimental variable seems to cause more
> trouble than it hurts.

No objections from me, I can remove if desired.

Thanks for your patience, I’m still finding my bearings on the C side.

Best,

--
Jacob S. Gordon
jacob.as.gordon <at> gmail.com
Please avoid sending me HTML emails and MS Office documents.
https://useplaintext.email/#etiquette

[v2-0001-Add-binary-format-specifications-b-and-B.patch (text/x-patch, attachment)]

[v2-0002-Use-printf-for-binary-formatting-when-available.patch (text/x-patch, attachment)]

Message #23 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Mattias Engdegård <mattias.engdegard <at> gmail.com>
To: Paul Eggert <eggert <at> cs.ucla.edu>,
 "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, 79990 <at> debbugs.gnu.org
Subject: bug#79990: [PATCH v1] Add binary format specifications '%b' and '%B' 
Date: Sun, 21 Dec 2025 11:37:27 +0100

Adding %b seems useful, thanks for proposing it.
(%B is just gold-plating, we can do without that.)

Not overly fond of using sub-normal bignums for printing fixnums but I have some sympathy for someone approaching `styled_format` for the first time. I've seen text editors that are shorter.

> Come to think of it, how about if we remove binary-as-unsigned instead?

Yes. I'll do it.

Message #26 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Mattias Engdegård <mattias.engdegard <at> gmail.com>
Cc: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Sun, 21 Dec 2025 10:02:08 -0800

[Message part 1 (text/plain, inline)]

On 2025-12-21 02:37, Mattias Engdegård wrote:
>> Come to think of it, how about if we remove binary-as-unsigned instead?
> Yes. I'll do it.

Thanks, but I had already done that privately, and as I added that 
experimental variable in the first place I might as well be the one to 
take the heat for removing it. I installed the attached.

> (%B is just gold-plating, we can do without that.)

Although it's indeed gold plating, %#b is also gold plating, and there 
is something to be said for supporting the exact same style of gold 
plating that the new C standard does.

[0001-Remove-binary-as-unsigned-bug-79990.patch (text/x-patch, attachment)]

Message #29 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Mattias Engdegård <mattias.engdegard <at> gmail.com>
To: Paul Eggert <eggert <at> cs.ucla.edu>
Cc: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Sun, 21 Dec 2025 21:11:29 +0100

21 dec. 2025 kl. 19.02 skrev Paul Eggert <eggert <at> cs.ucla.edu>:

>> Yes. I'll do it.
> 
> Thanks, but I had already done that privately, and as I added that experimental variable in the first place I might as well be the one to take the heat for removing it. I installed the attached.

Of course, and thank you for taking care of it all.
And for once, this is a variable the usage of which we have a sort of telemetry for.

> Although it's indeed gold plating, %#b is also gold plating, and there is something to be said for supporting the exact same style of gold plating that the new C standard does.

Oh well, the marginal cost is marginal.

Message #32 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Mattias Engdegård <mattias.engdegard <at> gmail.com>
Cc: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Sun, 21 Dec 2025 12:59:38 -0800

On 2025-12-21 12:11, Mattias Engdegård wrote:
> 21 dec. 2025 kl. 19.02 skrev Paul Eggert <eggert <at> cs.ucla.edu>:
> 
>>> Yes. I'll do it.
>>
>> Thanks, but I had already done that privately, and as I added that experimental variable in the first place I might as well be the one to take the heat for removing it. I installed the attached.
> 
> Of course, and thank you for taking care of it all.
> And for once, this is a variable the usage of which we have a sort of telemetry for.

Thanks for fixing the Texinfo typo in that patch's change to the manual.

However the following part of your fix wasn't correct because, for 
example, (read "-#xa") does not return -10. So I reverted that part.

> @@ -1225,11 +1225,10 @@ Formatting Strings
>  retrieve a copy of the formatted value, use a specification that lets
>  @code{read} reconstruct the value.  To format numbers in this
>  reversible way you can use @samp{%s} and @samp{%S}, to format just
> -integers you can also use @samp{%d}, and to format just nonnegative
> -integers you can also use @samp{#x%x} and @samp{#o%o}.  Other formats
> -may be problematic; for example, @samp{%d} and @samp{%g} can mishandle
> -NaNs and can lose precision and type, and @samp{#x%x} and @samp{#o%o}
> -can mishandle negative integers.  @xref{Input Functions}.
> +integers you can also use @samp{%d}, @samp{#x%x} and @samp{#o%o}.
> +Other formats do not guarantee preservation of all values; for example,
> +@samp{%g} can lose precision and type and does not preserve NaN or
> +infinity.  @xref{Input Functions}.
>  
>  The functions described in this section accept a fixed set of
>  specification characters.  The next section describes a function


Perhaps (read "-#xa") should return -10? But that would be a bigger change.

Message #35 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Mattias Engdegård <mattias.engdegard <at> gmail.com>
To: Paul Eggert <eggert <at> cs.ucla.edu>
Cc: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Sun, 21 Dec 2025 22:06:35 +0100

21 dec. 2025 kl. 21.59 skrev Paul Eggert <eggert <at> cs.ucla.edu>:

> However the following part of your fix wasn't correct because, for example, (read "-#xa") does not return -10.

Not sure what you mean. The text talks about "#x%x".

 (read (format "#x%x" -10)) -> -10

Message #38 received at 79990 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Mattias Engdegård <mattias.engdegard <at> gmail.com>
Cc: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>, 79990 <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v1] Add binary format specifications '%b' and
 '%B'
Date: Sun, 21 Dec 2025 13:24:17 -0800

On 2025-12-21 13:06, Mattias Engdegård wrote:
>   (read (format "#x%x" -10)) -> -10

Ouch, I completely misread the formats in the doc. Sorry about that; I 
re-reverted the doc change.

Message #43 received at 79990-done <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
Cc: Eli Zaretskii <eliz <at> gnu.org>, 79990-done <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v2] Add binary format specifications '%b' and
 '%B'
Date: Mon, 22 Dec 2025 00:30:44 -0800

[Message part 1 (text/plain, inline)]

On 2025-12-17 22:31, Jacob S. Gordon wrote:
> On 2025-12-12 13:09, Paul Eggert wrote:
>> Instead, I suggest changing styled_format's 'b' and 'B' processing
>> to convert fixnums to bignums and then "goto bignum_arg".
> 
> This makes sense, I’ve made this the default in v2.  AFAICT the
> existing bignum APIs assert numbers are in the fixnum range, so I had
> to expose arguments to skip them.  Is there a better way?

Yes, unfortunately it seems I led you down a bit of a wild-goose chase. 
Although that patch works, it complicates some bignum APIs to support a 
feature (bignums in fixnum range) that would be dicey elsewhere. On 
further thought, it's simpler to do the fixnum %b formatting by hand; 
it's a few lines of clearly-marked code, lines that we can remove a 
couple of decades from now.


>> On nicer platforms where plain printf works (e.g., glibc 2.35+) this
>> step can be omitted as an optimization
> 
> I took a stab at this in v2 with an autoconf macro, which checks a
> small program for the binary flags, e.g.
> 
>    #include <stdio.h>
>    int main() {
>        return printf("%b", 255) != 8;
>    }
> 
> This seemed like less of a maintenance burden than tracking down
> library compliance.  Is this on the right track?

For now let's just use the standard __STDC_VERSION_STDIO_H__ macro for 
library compliance, plus support glibc 2.35 through 2.42 even though 
they don't define  __STDC_VERSION_STDIO_H__. It's probably not worth the 
trouble of tracking down library compliance elsewhere (or of 
complicating and slowing down 'configure'), as this is mostly about code 
simplification in the future not about feature portability or 
performance in the present.

I adapted your patches with the above remarks in mind, installed the 
attached, and am boldly closing the bug report. Thanks for the bug 
report and the patches.

[0001-Add-binary-format-specifications-b-and-B.patch (text/x-patch, attachment)]

[0002-Simplify-new-b-B-code-many-years-from-now.patch (text/x-patch, attachment)]

Message #46 received at 79990-done <at> debbugs.gnu.org (full text, mbox):

From: "Jacob S. Gordon" <jacob.as.gordon <at> gmail.com>
To: Paul Eggert <eggert <at> cs.ucla.edu>
Cc: Eli Zaretskii <eliz <at> gnu.org>, mattias.engdegard <at> gmail.com,
 79990-done <at> debbugs.gnu.org
Subject: Re: bug#79990: [PATCH v2] Add binary format specifications '%b' and
 '%B'
Date: Mon, 22 Dec 2025 23:47:22 -0500

On 2025-12-22 03:30, Paul Eggert wrote:
> Although that patch works, it complicates some bignum APIs to
> support a feature (bignums in fixnum range) that would be dicey
> elsewhere. On further thought, it's simpler to do the fixnum %b
> formatting by hand; it's a few lines of clearly-marked code, lines
> that we can remove a couple of decades from now.

Oh much better, the manual formatting is simpler than I expected
(should have tried that!).

> For now let's just use the standard __STDC_VERSION_STDIO_H__ macro
> for library compliance, plus support glibc 2.35 through 2.42 even
> though they don't define __STDC_VERSION_STDIO_H__.

Sounds good.  It's still possible that __STDC_VERSION_STDIO_H__ >=
202311, but the optional %B spec does something different.  However, I
don’t have any definite example and this is a pretty niche feature, so
the current check is probably enough.

> I adapted your patches with the above remarks in mind, installed the
> attached, and am boldly closing the bug report.

Great, thanks for all your help (and Mattias for the follow-ups)!

Happy holidays,

-- 
Jacob S. Gordon
jacob.as.gordon <at> gmail.com
Please avoid sending me HTML emails and MS Office documents.
https://useplaintext.email/#etiquette

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.